banitag1

Support of buffering string output

3

### Is your feature request related to a problem? Please describe. Currently we are not able to store some strings output, the only option given are printing those immediately. We...

Add training common files for fbgemm inference mode

3

Summary: We need to reduce the binary size for sigrid_gpu and inference mode in fbgemm needed some more dependencies, so we created a training-common which can be included in the...

Use device index as zero if there is only one device

22

Summary: We are seeing that many benchmark use "cuda" instead of "cuda:0", so index is not properly set. This diff adds a simple change to set the target index as...

[PyTorch Pin Memory Allocator] Optimize the free list implementation and add lock sharding

9

Summary: This diff addresses the lock contention issue in free list implementation of CachingHost/Pinned allocator. We add a different data structure for free list and also add lock sharding based...